Exactly one week ago, Neeraj Arora, director of analytics at Farmers Insurance's Insight & Innovation group, took the stage at the 2008 Insurance & Technology Executive Summit and started his presentation on customer analytics thusly: "Knowing the customer is not something we can talk about in [the next 35 minutes] but we'll try."Arora's presentation was very in depth -- it made heavy usage of data contained within his slides to frame his discussion and engage the audience. So, similarly to how "knowing your customer" isn't something that be fully explained within the confines of a 35 minute presentation, perhaps Arora's presentation isn't something that can be explained within the confines of this blog post. That said, there are some key, high level takeaways from the presentation that I found particularly interesting.

We've written a lot about predictive analytics, customer analytics and data mining here, particularly over the past 12 months or so. One common theme that keeps coming up is, as my colleague Anthony O'Donnell put in a recent blog entry, "garbage in, garbage out" -- the concept that if you provide a predictive model with inaccurate or incomplete data, you'll end up with inaccurate or incomplete results.

After hearing Arora's presentation, I think there's another layer to this concept that could be just as important, and that is the importance of treating information gleaned from predictive models -- even accurate information -- with a healthy dose of skepticism.

Throughout Arora's discussion, he constantly wondered aloud about what the results of Farmers' models really meant. Is this model true for all customers? Are all customers reacting the same way to the same variable? Here was a numbers guy, who was constantly questioning the numbers.

"The questions you need to be asking yourself are 'Is everybody behaving the same way? Do all of us think about things in the same manner?' Most likely not," Arora explained.

The end goal of customer analytics, Arora said, is to truly understand how each individual customer is going to react or behave in a given instance. To achieve that though -- I inferred from Arora's talk -- it's important for an insurer to do its due diligence and drill down deep enough into the numbers and sort out what is truly useful from what is just simple correlation. "Garbage in, garbage out" is an important thing to keep in mind, but it doesn't end there. Even good information needs to be looked at critically.